Enhanced self organizing map with particle swarm optimization for classification problems

Abstract

Hybridization of Self Organizing Map (SOM) and Particle Swarm Optimization (PSO) is commonly implemented in clustering domain due to its capabilities in handling complex data characteristics. However, some of these hybrid architectures have weaknesses such as slow convergence time; always being trapped in the local minima and others. This study proposes a hybridization method by improving the Self Organizing Map (SOM) Lattice Structure with Particle Swarm Optimization (ESOMPSO) for solving classification problems. The enhancement of SOM lattice structure is implemented by introducing a new hexagon formulation for better mapping quality in data classification and labeling. The improvement of the SOM lattice structure using the proposed Enhanced SOM is implemented by optimizing the weights using PSO to obtain better output quality. The process is done in two stages: the first stage is conducted by training the weights using the Enhanced SOM, and the second stage is implemented by optimizing these weights with the PSO. The proposed method has been tested on various standard datasets. The comparisons are done on standard SOM, Enhanced SOM (ESOM), SOMPSO and ESOMPSO using various distance measurements. The performance of the proposed method is validated using classification accuracy and quantization error. The experiments have shown that ESOMPSO yields promising result with better average accuracy and quantization errors.